View
1
Download
0
Category
Preview:
Citation preview
Incremental Interactive Computation
Matthew Hammer hammer@cs.umd.edu
1Tuesday, October 7, 2014
Incremental Interactive Computation
Matthew Hammer hammer@cs.umd.edu
2Tuesday, October 7, 2014
Batch Computation vs Interactive Computation
InputOutputBatch Model
InteractiveModel
Input
Output
C
CInputOutput
U
No time,no dialogue
Dialogue in time
3Tuesday, October 7, 2014
Interaction is a DialogueCreates input
Demands Output
CU
InteractingUser
ReactiveComputation
Mutates input
Demands Output
TimeTime
4Tuesday, October 7, 2014
Elements of Interactive Dialogue
• User and system interact across time
• User mutates / changes input structure
• User demands / observes output structure
• The system maintains a correspondence between input and output structures
• I/O correspondence is computational
5Tuesday, October 7, 2014
Incremental Interactive Computation
6Tuesday, October 7, 2014
Incremental Interactive Computation
Input
Output
CInputOutput
U
Claim: Interesting interactive
systems consist of incremental
computations
7Tuesday, October 7, 2014
Example: SpreadsheetsInput
Structure Cell Formulae
IncrementalComputation
Formula evaluation
OutputStructure Cell values
8Tuesday, October 7, 2014
Example: Word processing
InputStructure
Document Content
IncrementalComputation
Spell-check each word
OutputStructure
Highlightmisspellings
9Tuesday, October 7, 2014
Example: ProgrammingInput
StructureText file
Computation Parse tokensOutput
StructureAST
InputStructure
ASTComputation Lexical scoping
OutputStructure
Def/use edges
InputStructure
AST + Def/useComputation Type inference
OutputStructure
Type errors
10Tuesday, October 7, 2014
Computing Incrementally1. Input changes are gradual
2. Output observation is limited
Full re-computation is often redundant
Full re-computation is often overly eager
11Tuesday, October 7, 2014
Computing Incrementally
Example:Spreadsheet Cells change slowly
Example:Word processing
Document and dictionary both change slowly
Example:Programming
Program changes slowly
1. Input changes are gradual
2. Output observation is limited
12Tuesday, October 7, 2014
Computing Incrementally
Example:Spreadsheet
One worksheet is active, others hidden
Example:Word processing
Viewport shows one or two pages
Example:Programming
Viewport shows one file, module or function
1. Input changes are gradual
2. Output observation is limited
13Tuesday, October 7, 2014
AdaptonProgramming Abstractionsfor Incremental Interaction
14Tuesday, October 7, 2014
Adapton Programming Abstractions
•Mutable references:
• Hold changing input structure
• Lazy thunks:
• Demand-driven computations
• Output structure
15Tuesday, October 7, 2014
max
1 2 3 4
+ min
+ *
max min
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
16Tuesday, October 7, 2014
max
1 2 3 4
+ min
+ *
max min
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
17Tuesday, October 7, 2014
max2
1 2 3 4
+5
min
+7
*
max7
min
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
18Tuesday, October 7, 2014
max2
1 2 3 4
+5
min
+7
*
max7
min
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs Switch Demand:
19Tuesday, October 7, 2014
max2
1 2 3 4
+5
min3
+7
*15
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs Switch Demand:
20Tuesday, October 7, 2014
max2
1 2 3 4
+5
min3
+7
*15
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
Sharing
21Tuesday, October 7, 2014
max2
1 2 3 4
+5
min3
+7
*15
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
Input Change:
22Tuesday, October 7, 2014
max2
1 10 3 4
+5
min3
+7
*15
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs
Some previous results are affected
Input Change:
23Tuesday, October 7, 2014
max2
1 10 3 4
+5
min3
+7
*15
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs Demand new output:
24Tuesday, October 7, 2014
max2
1 10 3 4
+13
min3
+7
*26
max7
min3
Mutablereferences
IncrementalComputation
(thunks)
Demand-drivenOutputs Demand new output:
25Tuesday, October 7, 2014
Lazy Structures• Spreadsheet example:
• Each thunk returns a single number
• Lazy Lists:
• Each thunk returns `Nil or `Cons
• `Cons holds head value and a thunk tail
• Laziness can be applied to trees, graphs, and essentially any other data structure
• Inputs: Special thunks are mutable
26Tuesday, October 7, 2014
Background: Lazy Liststype 'a thunk = unit -> 'a
let force : 'a thunk -> 'a = fun t -> t () type 'a lzlist = [ | `Nil | `Cons of 'a * ('a lzlist thunk) ]
let rec from_list l = match l with | [] -> `Nil | h::t -> `Cons(h,fun()->from_list t)
Apply unit argument
27Tuesday, October 7, 2014
type 'a lzlist = [ | `Nil | `Cons of 'a * ('a lzlist thunk) ]
let rec merge l1 l2 = function |l1, `Nil -> l1 |`Nil, l2 -> l2
|`Cons(h1,t1), `Cons(h2,t2) -> if h1 <= h2 then `Cons(h1, fun()->merge (force t1) l2) else `Cons(h2, fun()->merge l1 (force t2))
28Tuesday, October 7, 2014
Mergesort ExampleInput [3,5,8,2,1,7]
Singletons [ [3], [5], [8], [2], [1], [7] ]
Merge #1 [ [3,5], [2,8], [1,7] ]
Merge #2 [ [3,5], [1,2,7,8] ]
Merge #3 [ [3,5], [1,2,7,8] ]
Merge #4 [ [1, 2, 3, 5, 7, 8 ] ]
Flatten [1, 2, 3, 5, 7, 8 ]
29Tuesday, October 7, 2014
Course Project
30Tuesday, October 7, 2014
Project: Interactive Program Analysis
• Assume: Interactive “Structure Editor”
• Programmer manipulates AST directly
• Learn: Adapton IC framework
• Build: Incremental Program Analysis
• Example: Use/def information
• Example: Type Inference
• Example: Control-flow analysis
31Tuesday, October 7, 2014
Background: Structure Editors
• Philosophical claims:
• Programs consist of rich structure
• Rich interaction exposes this structure
• Example prototypes:
• Haskell -- http://www.youtube.com/watch?v=v2ypDcUM06U
• Citris -- http://www.youtube.com/watch?v=47UcOspbZ2k
• TouchDevelop -- http://www.youtube.com/watch?v=a6GRg2glKpc
32Tuesday, October 7, 2014
Recommended